Advanced Search

Publications > Journals > Journal of Clinical and Translational Hepatology > Article Full Text


Development and Validation of a Metabolic-related Prognostic Model for Hepatocellular Carcinoma

  • Junyu Huo,
  • Liqun Wu* and
  • Yunjin Zang
 Author information
Journal of Clinical and Translational Hepatology   2021;9(2):169-179

doi: 10.14218/JCTH.2020.00114


Background and Aims

Growing evidence suggests that metabolic-related genes have a significant impact on the occurrence and development of hepatocellular carcinoma (HCC). However, the prognostic value of metabolic-related genes for HCC has not been fully revealed.


mRNA sequencing and clinical data were obtained from The Cancer Genome Atlas and the GTEx Genotype-Tissue Expression comprehensive database. Differentially expressed metabolic-related genes in tumor tissues (n=374) and normal tissues (n=160) were identified by the Wilcoxon test. Time-dependent receiver operating characteristic curve analysis, univariate multivariate Cox regression analysis and Kaplan-Meier survival analysis were used to evaluate the predictive effectiveness and independence of the prognostic model. Two independent cohorts (International Cancer Genome Consortiums and GSE14520) were applied to verify the prognostic model.


Our study included a total of 793 patients with HCC. We constructed a risk score consisting of five metabolic-genes (BDH1, RRM2, CYP2C9, PLA2G7, and TXNRD1). For the overall survival rate, the low-risk group had a considerably higher rate than the high-risk group. Univariate and multivariate Cox regression analyses indicated that the risk score was an independent predictor for the prognosis of HCC.


We constructed and validated a novel prognostic model, which may provide support for the precise treatment of HCC.


Hepatocellular carcinoma, Metabolic, Prognostic, Signature


Hepatocellular carcinoma (HCC) is the second major cause of cancer-related death in the world.1,2 The prognosis of HCC is still not ideal, although related research has made great progress in recent years.1,3 This is mainly related to the high heterogeneity of HCC and the high diagnostic rate of advanced HCC,4–7 even with identical pathological types and clinical stages, patients’ individual responses to the same treatment can be diversified.8,9 It is worth noting that patients with HCC usually have a background of liver cirrhosis. Nevertheless, in clinical practice, which monitoring strategy is most effective for early tumor detection is uncertain.10 There is a pressing need to identify reliable biomarkers for the diagnosis and prognosis of HCC, to improve the survival of HCC.

In recent years, growing evidence has shown that the metabolic pattern of the cell cancerization process has changed significantly, which involves many aspects, such as glycolysis, the citric acid cycle, and oxidative phosphorylation of amino acids metabolism, fatty acid metabolism and nucleic acid metabolism, etc. This phenomenon is known as the reprogramming of energy metabolism of tumor cells, which is crucial for tumor growth.11,12 Some scholars have found that metabolic abnormalities are an important factor in the pathogenesis of HCC.6,7,13 Lee et al.13 noted that the gene expression levels involved in glycolysis and oxidative metabolism in HCC livers were much higher than those in normal livers, which was indeed relevant to an increased risk of liver cancer and may represent a potential target for the prevention of HCC. Gao et al.6 conducted a multidimensional proteomics study of 159 hepatitis B virus-positive liver and para cancer samples from patients in China, and found that most of the liver-specific metabolic pathway proteins (such as sugar dysplasia, detoxification, ammonia and urea metabolism) in liver tumors were significantly reduced; however, the key enzymes of cholesterol metabolism (SOAT1, SOAT2, etc.) and glutamine metabolism-related proteins (GLS and GLUD2) expressed in tumors were increased significantly, suggesting that hepato-specific metabolic pathways are reprogrammed in hepatitis B virus-associated HCC. Some scholars have also proposed that metabolic changes in the tumor microenvironment (TME) can inhibit antitumor immunity (such as immune cell infiltration) by producing immunosuppressive metabolites.14,15 However, there is still a lack of research on genes related to metabolism in predicting the prognosis of patients with liver cancer. Investigations into the metabolic genes of HCC are expected to open up new avenues for the treatment of HCC.

This research study established a risk score (RS) based upon the expression levels of five metabolism-related genes and analyzed the diverse clinicopathological features correlated with the new RS. The correlations of the RS with tumor immune cell infiltration were also evaluated.


Data collection and extraction of metabolic genes

The clinical data and mRNA expression profiles of patients with HCC were taken from The Cancer Genome Atlas (TCGA; https://portal.gdc.cancer.gov/ ), including 374 HCC samples and 50 normal samples; the Genotype-Tissue Expression project (GTEx; www.gtexportal.org ), including 110 normal samples; the International Cancer Genome Consortium (ICGC; https://icgc.org/ ), including 215 patients with HCC; and the Gene Expression Omnibus (GEO; https://www.ncbi.nlm.nih.gov/geo/ ) (GSE14520), including 235 patients with HCC. A total of 2,752 metabolic-related genes that encoded all the known human transporters and metabolic enzymes were obtained from a previously published paper,16 for subsequent analysis. The gene expression profiles obtained from different databases were normalized with the “combat” package in R software. The collected data were used in accordance with the data access strategies of TCGA, ICGC and GEO. All research processes and analyses were conducted in compliance with relevant regulations and guidelines. HCC clinical survival data and mRNA profile data were publicly available, and approval of the local ethics committee was not required.17

Identification of differentially expressed metabolic-related genes (DEMRGs)

Using the Wilcoxon method in the R package “limma” to detect differential genes related to metabolism in HCC and normal tissues, the results of log2 fold change >1 and false discovery rate <0.05 were regarded as significantly different. The “limma” and “heatmap” packages in the R software were used to form volcano and heat maps of DEMRGs.

Annotations of DEMRGs’ functions and pathways

This research applied the R package “cluster profile” for DEMRGs annotation (gene ontology [GO] and Kyoto Encyclopedia of Genes and Genomes [KEGG] pathway)18 to evaluate the underlying biological function of DEMRGs.

Identification of prognostic-related genes and construction of the prognostic model in the TCGA cohort

We used 343 patients (survival ≥1 month) from the TCGA dataset as the training cohort, to develop the prognostic model. During the building process of the prognostic model, we combined univariate Cox regression analysis, Lasso regression analysis, and multivariate Cox regression analyses. First, univariate Cox regression analysis was applied to screen DEMRGs associated with prognosis (p-value <0.001 considered significant).19,20 Next, the least absolute shrinkage and selection operator algorithm was utilized to avoid overfitting of the prognosis-related genes. During the process of this analysis, we subsampled the dataset 1,000 times and chose the genes that were repeated >900 times. A subselection of prognosis-related genes was determined by penalty parameter tuning performed via 10-fold cross-validation. Only genes with non-zero regression coefficients were retained for subsequent multivariate Cox regression analyses19–22 (Supplementary Fig. 1). The formula of the RS was as follows: RS = the sum of each multivariate Cox regression coefficient of mRNA multiplied by each normalized mRNA expression level. According to the median RS, patients were divided into two groups: the low-risk (LR) group and the high-risk (HR) group. Utilizing the Kaplan-Meier approach in the R-package “suvminer” produces survival curves, and the log-rank test was used to contrast discrepancies between the two groups. Using the time-dependent receiver operating characteristic (ROC) curve analysis feature in the R package, “survival ROC” aimed to evaluate the prognostic ability of the RS.

Independence validation of the prognostic model

Univariate and multivariate Cox regression analyses were applied to detect whether the RS was an independent prognostic predictor. A value of p<0.05 was statistically significant.

Internal validation of the prognostic model in the TCGA cohort

We divided the patients into several subgroups for internal validation according to their pathological features (including α-fetoprotein [AFP] level, vascular invasion, histological grade, AJCC-TNM stage, new tumor after initial treatment, and individual tumor status). The analysis of survival adopted the Kaplan-Meier method, and when the log-rank test detected a value of p<0.05, it was considered statistically significant.

External validation of the prognostic model using multiple independent cohorts

We calculated the RS of patients in the validation cohort (ICGC, GSE14520) using the same formula established by the TCGA cohort. The patients were separated into a LR group and a HR group based on the same cutoff value. Kaplan-Meier survival analysis, ROC curve analysis and univariate and multivariate Cox regression analysis were conducted as described above.

Correlation analysis between the RS and clinicopathology

We used the chi-square test to analyze the correlation between the RS and clinicopathology (including gender, age, AJCC-TNM stage, Barcelona Clinic Liver Cancer [BCLC] stage, Cancer of the Liver Italian Program [CLIP] stage, main tumor size, histologic grade, AFP, and vascular tumor cell type). A value of p<0.05 was statistically significant.

Correlation analysis between RS and tumor immune cell infiltration

The CIBERSORT method (using the characteristic matrix of 547 genes to express 22 types of infiltrating immune cells) was used to measure the infiltration ratio of immune cells as a number in tumor tissues, and the samples with p<0.05 were selected for subsequent analysis.

Gene set enrichment analysis (GSEA)

To further explore the internal mechanism of the prognostic model, we conducted GSEA on the LR and HR groups of the three independent cohorts to reveal the molecular biological characteristics of the LR and HR groups.


Functional enrichment analysis and survival analysis of DEMRGs

A total of 134 metabolic-related genes were differentially expressed in HCC tissues (n=374) compared with normal tissues (n=160) (Fig. 1A–B). GO enrichment analysis showed that the main functions of these differentially metabolized genes included small molecule catabolism, organic acid biosynthesis, sulfur compound metabolism, carboxylic acid biosynthesis, organic acid catabolism, fatty acid metabolism, carboxylic acid catabolism and other processes (Fig. 1C). The KEGG enrichment analysis identified these genes as being prevailingly related to chemical carcinogenesis, arachidonic acid metabolism, drug metabolism, glutathione metabolism, retinol metabolism, and carbon metabolism (Fig. 1D).

Identification and functional enrichment analysis of DEMRGs.
Fig. 1  Identification and functional enrichment analysis of DEMRGs.

(A–B) The heatmap and volcano plot of DMRGs. (C) GO enrichment analysis of DMRGs. (D) KEGG enrichment analysis of DMRGs.

Construction of the five-metabolic gene prognostic model

To facilitate the clinical application of our prognostic model, five metabolic-related genes were identified by Lasso-penalized Cox analysis to establish a predictive model. RS=(−0.02928*BDH1 normalized expression level)+(0.04763*RRM2 normalized expression level)+(−0.0018*CYP2C9 normalized expression level)+(0.0111*PLA2G7 normalized expression level)+(0.0111*TXNRD1 normalized expression level). The median RS (0.967) of the TCGA cohort is a critical value that divides all patients with HCC into HR and LR groups. This research applied disease-specific survival (DSS), overall survival (OS), progression-free survival (PFS) and disease-free survival (DFS) to compare the prognosis of patients with different risks. The Kaplan-Meier curve showed that compared with the HR group, the PFS, DFS, DSS, and OS of the LR group were remarkably higher (p<0.001) (Fig. 2A–D).

Kaplan-Meier survival analysis and time-dependent ROC analysis for (A) OS, (B) DSS, (C) DFS and (D) PFS.
Fig. 2  Kaplan-Meier survival analysis and time-dependent ROC analysis for (A) OS, (B) DSS, (C) DFS and (D) PFS.

When assessing the performance of the prognostic model by measuring the area over time under the ROC curve, the higher the area under the curve, the better was the model performance. The areas under the curve for the 1-year, 3-year, and 5-year PFS were 0.689, 0.621, and 0.683, respectively (Fig. 2D); the areas under the curve for the 1-year, 3-year, and 5-year DFS were 0.678, 0.615, and 0.691, respectively (Fig. 2C); the areas under the curve for the 1-year, 3-year, and 5-year DSS were 0.815, 0.738, and 0.674, respectively (Fig. 2B); and, the areas under the curve for the 1-year, 3-year, and 5-year OS were 0.8, 0.692, and 0.673, respectively (Fig. 2A). The RS was an independent prognostic indicator linked to PFS, DFS, DSS and OS, as presented by univariate and multivariate Cox regression analyses (Fig. 3A–D).

Forrest plot of the univariate and multivariate regression analysis regarding (A) OS, (B) DSS, (C) DFS and (D) PFS in the TCGA cohort.
Fig. 3  Forrest plot of the univariate and multivariate regression analysis regarding (A) OS, (B) DSS, (C) DFS and (D) PFS in the TCGA cohort.

Green represents univariate analysis, and red represents multivariate analysis.

Internal validation of the prognostic model in the TCGA cohort

We divided the patients into several subgroups for internal validation according to their pathological features, consistent with previous results. Compared to the LR group, the HR group patients’ OS rates were notably lower (Fig. 4A–F).

Internal validation in the TCGA cohort based on clinical features.
Fig. 4  Internal validation in the TCGA cohort based on clinical features.

(A) AFP. (B) Tumor status. (C) Histopathological grade. (D) New tumor event after initiate treatment. (E) AJCC-TNM stage. (F) Vascular tumor cell type.

External validation of the prognostic model in the ICGC and GSE14520 cohorts

Two independent datasets (ICGC, n=215; GSE14520, n=235) were used to test the prognostic value of the RS. The calculation formula of RS and the threshold value for dividing the HR and LR groups were consistent with that of the TCGA cohort. The HR patients’ OS was notably lower than that of LR patients in the two independent cohorts (Fig. 5A, D). The area under the ROC curve of the 1-year, 3-year and 5-year overall survival rates of the ICGC cohort and GSE14520 cohort were 0.750, 0.734, 0.829, 0.675, 0.671, and 0.673, respectively (Fig. 5A, D). The RS could be regarded as an independent prognostic indicator by univariate and multivariate Cox regression analyses (Fig. 5C, F). Due to the lack of relevant information about the Child/model for end-stage liver disease score, we cannot directly compare the prognostic value of the Child/model for end-stage liver disease score with the RS. However, by comparing the area under the curve values of the ROC curve between the RS and the traditional TNM staging system, we found that the RS had better performance in predicting prognosis (Supplementary Fig. 2).

External validation of the prognostic model in two independent cohorts.
Fig. 5  External validation of the prognostic model in two independent cohorts.

(A) Kaplan-Meier curve of OS and time-dependent ROC analysis in the ICGC cohort. (B) Heatmap of the five genes and the distribution of RS and the survival status of patients of the ICGC cohort. (C) Univariate and multivariate Cox regression analysis of the five-gene signature in the ICGC cohort (green represents univariate analysis, and red represents multivariate analysis). (D) Kaplan-Meier curve of OS and time-dependent ROC analysis in the GSE14520 cohort. (E) Heatmap of the five genes and the distribution of RS and the survival status of patients of the GSE14520 cohort. (F) Univariate and multivariate Cox regression analysis of the five-gene signature in the GSE14520 cohort (green represents univariate analysis, and red represents multivariate analysis).

Correlation of the prognostic model with clinicopathological characteristics

We performed chi-square tests on three independent cohorts (TCGA, ICGC, and GSE14520) and revealed that stage, grade, vascular tumor cell type, individual neoplasm status, main tumor size, and new tumor event after initial treatment concerned the RS of patients with HCC (Supplementary Tables 13).

Correlation analysis between the RS and tumor immune cell infiltration

The CIBERSORT algorithm was used to further analyze the infiltration degree of immune cell subtypes (samples were screened by p<0.05). The results showed that, compared with the LR group, the HR group had a markedly higher infiltration degree of M2 macrophages and a lower infiltration degree of M1 macrophages (Fig. 6A–C).

Relative proportion of 22 kinds of immune cell infiltration in HR and LR patients estimated by the CIBERSORT method.
Fig. 6  Relative proportion of 22 kinds of immune cell infiltration in HR and LR patients estimated by the CIBERSORT method.

(A) Barplot. (B) Heatmap. (C) Radar plot visualizing significantly different infiltration immune cells between HR and LR groups (p-value significant codes: 0≤ *** <0.001 ≤ ** < 0.01 ≤ * < 0.05).


As shown in Supplementary Fig. 3, the activity of metabolism-related pathways in the LR group was significantly stronger than that in the HR group, suggesting that we may be able to find new therapeutic strategies to improve the prognosis of HCC by targeting metabolic reprogramming of HCC.


Because HCC usually occurs in the context of cirrhosis, it has high morbidity, mortality, recurrence and heterogeneity,1–3 and poses a great threat to human health. With the worldwide application of next-generation gene sequencing technology, people have gradually realized that the prognosis of patients with HCC is not only dependent on the traditional clinical staging system but is also related to molecular genetic factors.5,7,23–25 Growing evidence has shown that metabolic reprogramming exerts a huge function in the emergence and growth of HCC.11,13 The prognostic value of some metabolic genes has been validated,6,26 but they are still numerically inadequate. Thus, there remains a pressing need to identify more biomarkers related to the prognosis of HCC.20

Compared with previous studies,18–20,26,27 this study highlighted the following aspects. First, in this study, mRNA data from TCGA, ICGC, GEO, and GTEx were integrated to study the prognostic value of metabolic-related genes in HCC. Second, we used three independent cohorts (TCGA, ICGC, and GSE14520) to construct and validate the prognostic model, making the conclusion more reliable. Third, we investigated the relationship between the prognostic model and tumor immune cell infiltration.

Patients were classified into a LR group and a HR group, based on a uniform cutoff (0.967). In both the training cohort (TCGA) and the validation cohort (ICGC and GSE14520), the OS of the HR group was significantly lower than that of the LR group. The area under the curve of the ROC curve showed that the RS had higher specificity and sensitivity for the prediction of prognosis. Clinical correlation analysis showed that patients with a HR score were found to be significantly correlated with higher tumor grade, larger main tumor size (>5 cm), older (age >65 years), vessel invasion, AFP >300 ng/mL, advanced BCLC stage (B–C), advanced CLIP stage (≥2) and advanced TNM stage (III–IV), and these results suggest the high-RS patients had a higher degree of malignancy. For patients with the same clinical features, the prognosis of the HR group patients was markedly worse than for those of the LR group, which highlights the importance of our RS establishment because it can better reflect the heterogeneity of patients compared with the traditional clinical stage.

It has been reported that abnormal metabolism of purines (such as abnormal elevation of uric acid) is related to the appearance and growth of different malignant tumors, such as colorectal cancer metastasis, non-small cell lung cancer brain metastasis, and the prognosis of pancreatic cancer, etc.,28–33 but there have been few reports on HCC. Immune cells are the main non-tumor components in the TME, and earlier studies have suggested that the infiltration of immune cells (neutrophils, dendritic cells, macrophages, etc.) in tumors has a close relationship with unadvanced HCC prognosis.34–39 Macrophages are the most numerous in tumor tissues and have the most significant regulatory effect on tumors. As such, they are called tumor-associated macrophages. Studies have found that M1-type macrophages can recognize tumor antigens, and phagocytose or kill tumor cells. Type II interferon (interferon-γ) is a classic inducer of M1 macrophage polarization and tumor cell killing.36,38 M2-type macrophages inhibit the activation and proliferation of T cells and natural killer cells by producing interleukin-10, transforming growth factor-β and prostaglandin E2 (prostaglandin E-2, PGE-2), and induce immune tolerance of tumor cells, thus promoting the proliferation, invasion and metastasis of tumor cells.36 In this study, it was found that the HR group had a higher infiltration degree of M2 macrophages and a lower degree of M1 macrophages than the LR group. These results suggested that this model could be used as an effective predictor of immune cell infiltration.

At present, obesity and metabolic diseases have become important factors that induce liver cancer.40 The beige fat cells in the body have come under scrutiny for their ability to burn energy to prevent obesity.41,42 Wang et al.43 found that catabolic metabolism of n-hydroxybutyrate mediated by 3-hydroxybutyrate dehydrogenase (BDH1) is an important step in the formation of beige fat cells in the body. Martinez-Outschoorn et al.44,45 found that BDH1 is preferentially expressed in breast tumor mesenchymal cells and that overexpression of BDH1 can promote the growth of cancer cells by generating fibroblasts to drive increased mitochondrial synthesis. Saraon et al.46 also found significant up-regulation of BDH1 expression in prostate cancer tissues. The current study discovered that, compared with normal tissues, BDH1 expression in tumor tissues was markedly down-regulated (Supplementary Fig. 4). Moreover, the high expression of BDH1 in tumor tissues was correlated with better OS and earlier TNM stage and a maximum tumor diameter of ≤5 cm and AFP of ≤300 ng/mL (Supplementary Figs. 58). Whether the mechanism is related to the previously reported promotion of beige fat cell formation in vivo deserves further study.

The ribonucleotide reductase m2 RRM2 has been confirmed repeatedly to have a relationship with HCC prognosis in recent years.47–49 Kosakowska et al.50 found a reduction in RRM2 remarkably suppressed HCC cell proliferation, and that RRM2 catalyzes the conversion of ribonucleoside 5′-diphosphate into a corresponding 2′-deoxyribonucleotide, and since this reaction is a rate-limiting step in DNA synthesis, RRM2 has been identified as a new target for cancer therapy.51,52 Additionally, this study discovered that poor OS was responsible for high RRM2 expression and predicted low tumor differentiation (Supplementary Figs. 58).

The cytochrome P450 system of the liver plays an important role in drug metabolism. The CYP2 family is the largest family of the CYP450 enzyme family, among which CYP2C9 is one of the most important subtypes.53 Nebert et al.54 once reported that the expression of CYP can affect the production of arachidonic acid-derived molecules and change various downstream signal transduction pathways, thus causing cell cancerization. Yan et al.’s55 study found that the expression level of CYP was significantly destroyed during the cancer process, while the activity of CYP was highly correlated with the expression level of the protein. The current study found that the expression level of CYP2C9 in tumor tissue was remarkably lower than that in normal tissue (Supplementary Fig. 4), and patients with high CYP2C9 expression in tumor tissues had a better prognosis (Supplementary Figs. 58). In addition, the reduced CYP2C9 expression level had a significant relationship with higher TNM stage and higher BCLC stage, a maximum tumor diameter of >5 cm, AFP of >300 ng/mL and vessel invasion (Supplementary Figs. 58). The results were similar to those of CYP4A11 in the study by Eun et al.56 These findings indicate that the high CYP2C9 expression is likely to be a favorable signal for the prognosis of HCC (Supplementary Figs. 58). Clinically, we can consider using CYP2C9 as a therapeutic target to further improve the prognosis of HCC.

The platelet-activating factor acetylhydrolase PLA2G7 is an effective proinflammatory and anti-inflammatory molecule involved in a variety of inflammatory processes.57 Nair et al.58 found that PLA2G7 expression was significantly up-regulated in fat cell precursors in obese individuals. Hou et al.59 and Hoffmann et al.60 identified PLA2G7 as a risk factor for cardiovascular disease. The current study found that compared with the level in normal tissues, the PLA2G7 expression level in tumors was remarkably higher (Supplementary Fig. 4), and the high expression level of PLA2G7 was significantly correlated with AFP >300 ng/mL (Supplementary Figs. 58), so PLA2G7 may be a new diagnostic marker for HCC.

Increasing evidence shows that oxidative stress caused by the destruction of the reduction-oxidation system is closely related to the occurrence of liver cancer.61–63 The thioredoxin reductase 1 TXNRD1, as a member of the thioredoxin system, is essential for maintaining the balance of the redox state in cells.62,64 This study found that TXNRD1 was significantly correlated with poor OS and higher TNM staging (Supplementary Figs. 58), which was similar to the report of Fu et al.65 and Lee et al.66 Therefore, TXNRD1 may be a biomarker with important prognostic value for HCC.

Targeted sequencing based on five metabolic genes can undoubtedly significantly reduce the cost of sequencing, but there are some limitations because our research results are mainly based on the description of the phenomenon, and we need to explore its mechanism through experiments. HCC is a complex disease caused by multiple mechanisms, not just metabolic disorders. Although we made full use of data resources, the lack of some clinical data will cause inevitable limitations, for example, the adjuvant treatment methods patients receive, such as chemotherapy, targeted therapy, and immunotherapy, comorbidities of patients and whether patients have underlying cirrhosis, because these factors have a significant impact on the clinical outcome. This study was retrospective and needs to be improved upon and verified in future multicenter prospective studies.


This research established and verified a reliable prognostic model for HCC patients. The five metabolic genes in the model may be promising targets for the precise treatment of HCC. Therefore, it is likely to have influential potential for clinical practice in the near future.

Supporting information

Supplementary Fig. 1

Building process of the prognostic model.

(A) Univariate Cox regression analysis. (B) Lasso regression analysis. (C) Multivariate Cox regression analysis.


Supplementary Fig. 2

Comparison of prognostic performance between RS and TNM staging.


Supplementary Fig. 3

GSEA between the different risk groups.

(A) TCGA cohort. (B) ICGC cohort. (C) GSE14520 cohort.


Supplementary Fig. 4

Validation of the expression pattern of the five gene signature.

(A) ICGC cohort (240 tumor tissues vs. 202 normal tissues). (B) GSE14520 cohort (244 tumor tissues vs. 110 normal tissues).


Supplementary Fig. 5

Correlation analysis between gene expression and clinical features in the TCGA cohort.


Supplementary Fig. 6

Correlation analysis between gene expression and clinical features in the ICGC cohort.


Supplementary Fig. 7

Correlation analysis between gene expression and clinical features in the GSE14520 cohort.


Supplementary Fig. 8

Correlation of gene expression with OS and recurrence-free survival.

This image was downloaded from the online Kaplan-Meier plotter.


Supplementary Table 1

Chi-square test of the relation between RS and clinical features in TCGA.


Supplementary Table 2

Chi-square test of the relation between RS and clinical features in ICGC.


Supplementary Table 3

Chi-square test of the relation between RS and clinical features in GSE14520.






glutamic pyruvic transaminase


Barcelona Clinic Liver Cancer


3-hydroxybutyrate dehydrogenase


Cancer of the Liver Italian Program


differentially expressed metabolic-related genes


disease-free survival


disease-specific survival


Gene Expression Ominibus


Gene Ontology


gene set enrichment analysis


Genotype-Tissue Expression comprehensive database


hepatocellular carcinoma




International Cancer Genome Consortium


Kyoto Encyclopedia of Genes and Genomes




overall survival


progression-free survival


receiver operating characteristic


risk score


The Cancer Genome Atlas


tumor microenvironment



None to declare.

Conflict of interest

The authors have no conflict of interests related to this publication.

Authors’ contributions

Designed this study (JH, LW), performed the data analyses and wrote the manuscript (JH), revision of the manuscript (JH, LW, YZ), reviewed the final version of the manuscript (JH, LW, YZ).


  1. Craig AJ, von Felden J, Garcia-Lezana T, Sarcognato S, Villanueva A. Tumour evolution in hepatocellular carcinoma. Nat Rev Gastroenterol Hepatol 2020;17:139-152 View Article
  2. Yang JD, Hainaut P, Gores GJ, Amadou A, Plymoth A, Roberts LR. A global view of hepatocellular carcinoma: trends, risk, prevention and management. Nat Rev Gastroenterol Hepatol 2019;16:589-604 View Article
  3. Zheng R, Qu C, Zhang S, Zeng H, Sun K, Gu X, et al. Liver cancer incidence and mortality in China: Temporal trends and projections to 2030. Chin J Cancer Res 2018;30:571-579 View Article
  4. Brunner SF, Roberts ND, Wylie LA, Moore L, Aitken SJ, Davies SE, et al. Somatic mutations and clonal dynamics in healthy and cirrhotic human liver. Nature 2019;574:538-542 View Article
  5. Ding X, He M, Chan AWH, Song QX, Sze SC, Chen H, et al. Genomic and epigenomic features of primary and recurrent hepatocellular carcinomas. Gastroenterology 2019;157:1630-1645.e6 View Article
  6. Gao Q, Zhu H, Dong L, Shi W, Chen R, Song Z, et al. Integrated proteogenomic characterization of HBV-related hepatocellular carcinoma. Cell 2019;179:561-577.e22 View Article
  7. Nakagawa H, Fujita M, Fujimoto A. Genome sequencing analysis of liver cancer for precision medicine. Semin Cancer Biol 2019;55:120-127 View Article
  8. Hage C, Hoves S, Strauss L, Bissinger S, Prinz Y, Pöschinger T, et al. Sorafenib induces pyroptosis in macrophages and triggers natural killer cell-mediated cytotoxicity against hepatocellular carcinoma. Hepatology 2019;70:1280-1297 View Article
  9. Bruix J, da Fonseca LG, Reig M. Insights into the success and failure of systemic therapy for hepatocellular carcinoma. Nat Rev Gastroenterol Hepatol 2019;16:617-630 View Article
  10. Tzartzeva K, Obi J, Rich NE, Parikh ND, Marrero JA, Yopp A, et al. Surveillance imaging and alpha fetoprotein for early detection of hepatocellular carcinoma in patients with cirrhosis: A meta-analysis. Gastroenterology 2018;154:1706-1718.e1 View Article
  11. Hanahan D, Weinberg RA. Hallmarks of cancer: the next generation. Cell 2011;144:646-674 View Article
  12. Ward PS, Thompson CB. Metabolic reprogramming: a cancer hallmark even warburg did not anticipate. Cancer Cell 2012;21:297-308 View Article
  13. Lee NCW, Carella MA, Papa S, Bubici C. High expression of glycolytic genes in cirrhosis correlates with the risk of developing liver cancer. Front Cell Dev Biol 2018;6:138 View Article
  14. Li X, Wenes M, Romero P, Huang SC, Fendt SM, Ho PC. Navigating metabolic pathways to enhance antitumour immunity and immunotherapy. Nat Rev Clin Oncol 2019;16:425-441 View Article
  15. Xia S, Pan Y, Liang Y, Xu J, Cai X. The microenvironmental and metabolic aspects of sorafenib resistance in hepatocellular carcinoma. EBioMedicine 2020;51:102610 View Article
  16. Possemato R, Marks KM, Shaul YD, Pacold ME, Kim D, Birsoy K, et al. Functional genomics reveal that the serine synthesis pathway is essential in breast cancer. Nature 2011;476:346-350 View Article
  17. Long J, Wang A, Bai Y, Lin J, Yang X, Wang D, et al. Development and validation of a TP53-associated immune prognostic model for hepatocellular carcinoma. EBioMedicine 2019;42:363-374 View Article
  18. Liu F, Liao Z, Song J, Yuan C, Liu Y, Zhang H, et al. Genome-wide screening diagnostic biomarkers and the construction of prognostic model of hepatocellular carcinoma. J Cell Biochem 2020;121:2582-2594 View Article
  19. Wang Z, Zhu J, Liu Y, Liu C, Wang W, Chen F, et al. Development and validation of a novel immune-related prognostic model in hepatocellular carcinoma. J Transl Med 2020;18:67 View Article
  20. Li W, Lu J, Ma Z, Zhao J, Liu J. An integrated model based on a six-gene signature predicts overall survival in patients with hepatocellular carcinoma. Front Genet 2020;10:1323 View Article
  21. Huo J, Wu L, Zang Y. A prognostic model of 15 immune-related gene pairs associated with tumor mutation burden for hepatocellular carcinoma. Front Mol Biosci 2020;7:581354 View Article
  22. Huo J, Wu L, Zang Y. A robust nine-gene prognostic signature associated with tumour doubling time for hepatocellular carcinoma. Life Sci 2020;260:118396 View Article
  23. Huitzil-Melendez FD, Capanu M, O’Reilly EM, Duffy A, Gansukh B, Saltz LL, et al. Advanced hepatocellular carcinoma: which staging systems best predict prognosis?. J Clin Oncol 2010;28:2889-2895 View Article
  24. Nault JC, Martin Y, Caruso S, Hirsch TZ, Bayard Q, Calderaro J, et al. Clinical impact of genomic diversity from early to advanced hepatocellular carcinoma. Hepatology 2020;71:164-182 View Article
  25. Zhang Q, Lou Y, Yang J, Wang J, Feng J, Zhao Y, et al. Integrated multiomic analysis reveals comprehensive tumour heterogeneity and novel immunophenotypic classification in hepatocellular carcinomas. Gut 2019;68:2019-2031 View Article
  26. Liu GM, Zeng HD, Zhang CY, Xu JW. Identification of a six-gene signature predicting overall survival for hepatocellular carcinoma. Cancer Cell Int 2019;19:138 View Article
  27. Liu GM, Xie WX, Zhang CY, Xu JW. Identification of a four-gene metabolic signature predicting overall survival for hepatocellular carcinoma. J Cell Physiol 2020;235:1624-1636 View Article
  28. Stotz M, Szkandera J, Seidel J, Stojakovic T, Samonigg H, Reitz D, et al. Evaluation of uric acid as a prognostic blood-based marker in a large cohort of pancreatic cancer patients. PLoS One 2014;9:e104730 View Article
  29. Tanriverdi O, Cokmert S, Oktay E, Pilanci KN, Menekse S, Kocar M, et al. Prognostic significance of the baseline serum uric acid level in non-small cell lung cancer patients treated with first-line chemotherapy: a study of the Turkish Descriptive Oncological Researches Group. Med Oncol 2014;31:217 View Article
  30. Chen YF, Li Q, Chen DT, Pan JH, Chen YH, Wen ZS, et al. Prognostic value of pre-operative serum uric acid levels in esophageal squamous cell carcinoma patients who undergo R0 esophagectomy. Cancer Biomark 2016;17:89-96 View Article
  31. Selcukbiricik F, Kanbay M, Solak Y, Bilici A, Kanıtez M, Balık E, et al. Serum uric acid as a surrogate marker of favorable response to bevacizumab treatment in patients with metastatic colon cancer. Clin Transl Oncol 2016;18:1082-1087 View Article
  32. Yuan C, Xu XH, Wang XL, Xu L, Chen Z, Li YQ. Relationship between serum uric acid and metastatic and nonmetastatic rectal cancer patients with undergoing no chemotherapy. Medicine (Baltimore) 2016;95:e5463 View Article
  33. Cetin AO, Omar M, Calp S, Tunca H, Yimaz N, Ozseker B, et al. Hyperuricemia at the time of diagnosis is a factor for poor prognosis in patients with stage II and III colorectal cancer (uric acid and colorectal cancer). Asian Pac J Cancer Prev 2017;18:485-490 View Article
  34. Kuang DM, Zhao Q, Wu Y, Peng C, Wang J, Xu Z, et al. Peritumoral neutrophils link inflammatory response to disease progression by fostering angiogenesis in hepatocellular carcinoma. J Hepatol 2011;54:948-955 View Article
  35. Mantovani A, Cassatella MA, Costantini C, Jaillon S. Neutrophils in the activation and regulation of innate and adaptive immunity. Nat Rev Immunol 2011;11:519-531 View Article
  36. Wu SD, Ma YS, Fang Y, Liu LL, Fu D, Shen XZ. Role of the microenvironment in hepatocellular carcinoma development and progression. Cancer Treat Rev 2012;38:218-225 View Article
  37. Mantovani A, Marchesi F, Malesci A, Laghi L, Allavena P. Tumour-associated macrophages as treatment targets in oncology. Nat Rev Clin Oncol 2017;14:399-416 View Article
  38. Degroote H, Van Dierendonck A, Geerts A, Van Vlierberghe H, Devisscher L. Preclinical and clinical therapeutic strategies affecting tumor-associated macrophages in hepatocellular carcinoma. J Immunol Res 2018;2018:7819520 View Article
  39. Zhou ZJ, Xin HY, Li J, Hu ZQ, Luo CB, Zhou SL. Intratumoral plasmacytoid dendritic cells as a poor prognostic factor for hepatocellular carcinoma following curative resection. Cancer Immunol Immunother 2019;68:1223-1233 View Article
  40. Grohmann M, Wiede F, Dodd GT, Gurzov EN, Ooi GJ, Butt T, et al. Obesity drives STAT-1-dependent NASH and STAT-3-dependent HCC. Cell 2018;175:1289-1306.e20 View Article
  41. Kajimura S, Seale P, Tomaru T, Erdjument-Bromage H, Cooper MP, Ruas JL, et al. Regulation of the brown and white fat gene programs through a PRDM16/CtBP transcriptional complex. Genes Dev 2008;22:1397-1409 View Article
  42. Harms M, Seale P. Brown and beige fat: development, function and therapeutic potential. Nat Med 2013;19:1252-1263 View Article
  43. Wang W, Ishibashi J, Trefely S, Shao M, Cowan AJ, Sakers A, et al. A PRDM16-driven metabolic signal from adipocytes regulates precursor cell fate. Cell Metab 2019;30:174-189.e5 View Article
  44. Martinez-Outschoorn UE, Lin Z, Whitaker-Menezes D, Howell A, Lisanti MP, Sotgia F. Ketone bodies and two-compartment tumor metabolism: stromal ketone production fuels mitochondrial biogenesis in epithelial cancer cells. Cell Cycle 2012;11:3956-3963 View Article
  45. Martinez-Outschoorn UE, Lin Z, Whitaker-Menezes D, Howell A, Sotgia F, Lisanti MP. Ketone body utilization drives tumor growth and metastasis. Cell Cycle 2012;11:3964-3971 View Article
  46. Saraon P, Cretu D, Musrap N, Karagiannis GS, Batruch I, Drabovich AP, et al. Quantitative proteomics reveals that enzymes of the ketogenic pathway are associated with prostate cancer progression. Mol Cell Proteomics 2013;12:1589-1601 View Article
  47. Wu M, Liu Z, Li X, Zhang A, Lin D, Li N. Analysis of potential key genes in very early hepatocellular carcinoma. World J Surg Oncol 2019;17:77 View Article
  48. Yue C, Ren Y, Ge H, Liang C, Xu Y, Li G, et al. Comprehensive analysis of potential prognostic genes for the construction of a competing endogenous RNA regulatory network in hepatocellular carcinoma. Onco Targets Ther 2019;12:561-576 View Article
  49. Zhou Z, Li Y, Hao H, Wang Y, Zhou Z, Wang Z, et al. Screening hub genes as prognostic biomarkers of hepatocellular carcinoma by bioinformatics analysis. Cell Transplant 2019;28:76S-86S View Article
  50. Kosakowska-Cholody T, Cholody WM, Hariprakasha HK, Monks A, Kar S, Wang M, et al. Growth inhibition of hepatocellular carcinoma cells in vitro and in vivo by the 8-methoxy analog of WMC79. Cancer Chemother Pharmacol 2009;63:769-778 View Article
  51. Satow R, Shitashige M, Kanai Y, Takeshita F, Ojima H, Jigami T, et al. Combined functional genome survey of therapeutic targets for hepatocellular carcinoma. Clin Cancer Res 2010;16:2518-2528 View Article
  52. Gao J, Chen H, Yu Y, Song J, Song H, Su X, et al. Inhibition of hepatocellular carcinoma growth using immunoliposomes for co-delivery of adriamycin and ribonucleotide reductase M2 siRNA. Biomaterials 2013;34:10084-10098 View Article
  53. Korobkova EA. Effect of natural polyphenols on CYP metabolism: Implications for diseases. Chem Res Toxicol 2015;28:1359-1390 View Article
  54. Nebert DW, Dalton TP. The role of cytochrome P450 enzymes in endogenous signalling pathways and environmental carcinogenesis. Nat Rev Cancer 2006;6:947-960 View Article
  55. Yan T, Lu L, Xie C, Chen J, Peng X, Zhu L, et al. Severely impaired and dysregulated cytochrome P450 expression and activities in hepatocellular carcinoma: Implications for personalized treatment in patients. Mol Cancer Ther 2015;14:2874-2886 View Article
  56. Eun HS, Cho SY, Lee BS, Kim S, Song IS, Chun K, et al. Cytochrome P450 4A11 expression in tumor cells: A favorable prognostic factor for hepatocellular carcinoma patients. J Gastroenterol Hepatol 2019;34:224-233 View Article
  57. Sutton BS, Crosslin DR, Shah SH, Nelson SC, Bassil A, Hale AB, et al. Comprehensive genetic analysis of the platelet activating factor acetylhydrolase (PLA2G7) gene and cardiovascular disease in case-control and family datasets. Hum Mol Genet 2008;17:1318-1328 View Article
  58. Nair S, Lee YH, Rousseau E, Cam M, Tataranni PA, Baier LJ, et al. Increased expression of inflammation-related genes in cultured preadipocytes/stromal vascular cells from obese compared with non-obese Pima Indians. Diabetologia 2005;48:1784-1788 View Article
  59. Hou L, Chen S, Yu H, Lu X, Chen J, Wang L, et al. Associations of PLA2G7 gene polymorphisms with plasma lipoprotein-associated phospholipase A2 activity and coronary heart disease in a Chinese Han population: the Beijing atherosclerosis study. Hum Genet 2009;125:11-20 View Article
  60. Hoffmann MM, Winkler K, Renner W, Winkelmann BR, Seelhorst U, Wellnitz B, et al. Genetic variants and haplotypes of lipoprotein associated phospholipase A2 and their influence on cardiovascular disease (The Ludwigshafen Risk and Cardiovascular Health Study). J Thromb Haemost 2009;7:41-48 View Article
  61. Reuter S, Gupta SC, Chaturvedi MM, Aggarwal BB. Oxidative stress, inflammation, and cancer: how are they linked?. Free Radic Biol Med 2010;49:1603-1616 View Article
  62. Cho SY, Kim S, Son MJ, Rou WS, Kim SH, Eun HS, et al. Clinical significance of the thioredoxin system and thioredoxin-domain-containing protein family in hepatocellular carcinoma. Dig Dis Sci 2019;64:123-136 View Article
  63. Casas-Grajales S, Muriel P. Antioxidants in liver health. World J Gastrointest Pharmacol Ther 2015;6:59-72 View Article
  64. Mahmood DF, Abderrazak A, El Hadri K, Simmet T, Rouis M. The thioredoxin system as a therapeutic target in human health and disease. Antioxid Redox Signal 2013;19:1266-1303 View Article
  65. Fu B, Meng W, Zeng X, Zhao H, Liu W, Zhang T. TXNRD1 is an unfavorable prognostic factor for patients with hepatocellular carcinoma. Biomed Res Int 2017;2017:4698167 View Article
  66. Lee D, Xu IM, Chiu DK, Leibold J, Tse AP, Bao MH, et al. Induction of oxidative stress through inhibition of thioredoxin reductase 1 is an effective therapeutic approach for hepatocellular carcinoma. Hepatology 2019;69:1768-1786 View Article
  • Journal of Clinical and Translational Hepatology
  • pISSN 2225-0719
  • eISSN 2310-8819
Back to Top

Development and Validation of a Metabolic-related Prognostic Model for Hepatocellular Carcinoma

Junyu Huo, Liqun Wu, Yunjin Zang
  • Reset Zoom
  • Download TIFF